Introduction:
For decades, electricity in America was treated as invisible infrastructure. Cheap. Reliable. Politically boring. The lines running from the substation to the socket were background noise in the national conversation about growth, innovation, and prosperity. Households paid their utility bills without much scrutiny, and corporations negotiated favorable industrial rates with minimal public accountability. Electricity was a utility in the truest sense of the word: something everyone needed, something no one talked about, and something that was presumed, like gravity, to simply exist.
That era is ending.
The rise of artificial intelligence has done something no previous technology wave quite managed: it has converted electricity from a passive input into an active strategic resource. The construction of hyperscale AI data centers—those vast, energy-voracious complexes operated by Microsoft, Amazon, Alphabet, Meta, and Oracle—has introduced a new category of industrial power consumer whose appetite for electricity is measured not in megawatts but in gigawatts, not in monthly contracts but in multi-decade power purchase agreements. These facilities do not merely consume electricity; they reshape the economics, geography, and politics of the entire grid around them.
This is not merely an engineering story. It is a political economy story. It is about who gets priority access to electricity. It is about whether ordinary residents subsidize infrastructure expansion designed primarily for trillion-dollar corporations. It is about whether the AI economy socializes cost while privatizing upside. And as voters in Virginia and New Jersey demonstrated in November 2025, it is becoming one of the most consequential political issues of the era, one that shows every sign of metastasizing into a defining 2026 midterm election theme.
The signal that this is no longer a speculative concern arrived on May 4, 2026, when the North American Electric Reliability Corporation (NERC) issued a rare Level 3 “Essential Actions” Alert—its highest-urgency notification—in response to a documented pattern of customer-initiated load reductions in which 1,000 or more megawatts of computational load dropped off the bulk power system in seconds.1 These events, occurring across both the Eastern and Texas interconnections since 2022, exposed a fundamental fragility: the grid’s bulk power system was not designed to absorb the sudden, massive disconnections that occur when AI data centers automatically trip offline to protect their sensitive computing equipment from electrical disturbance.
The NERC alert is not merely a technical bulletin. It is a declaration that the American grid has entered structurally new territory. The phrase used in the alert—“large computational loads”—is bureaucratic language for a seismic transformation in the national electricity system, one that carries with it profound implications for the 130 million residential ratepayers who have no negotiating power at the utility table.2
This paper examines that transformation through five interconnected lenses: the political economy of electricity and its implications for the 2026 midterm elections; the technical realities of national grid stress as documented by NERC, the IEA, and independent regulators; the state-by-state fragmentation of the American electricity system under the weight of hyperscale demand; the nuclear revival as both a potential solution and a potential accelerant of private power inequality; and, finally, the strategic lessons that policymakers, regulators, and citizens must internalize if the social contract surrounding electricity is to survive the AI transition.
The paper takes its title not from an engineering term but from an emerging political metaphor. “Power Grid Constraints” is the phrase utilities and grid operators use to describe the limits of the infrastructure. But it is fast becoming the phrase that voters will use to describe the limits of fairness in the AI economy.
Section 1: The Politics of Electricity: Why Power Grid Constraints Could Become a Midterm Issue
Electricity has historically been among the least politically volatile of infrastructure costs. Unlike gasoline, whose price is posted in enormous digits on roadside signs and whose fluctuations track the news cycle almost in real time, electricity bills arrive monthly, are denominated in units most consumers cannot intuitively understand—the kilowatt-hour—and have moved slowly enough over decades to remain beneath the threshold of political mobilization. That structural invisibility is now collapsing, and the speed of its collapse is directly correlated with the speed of AI data center construction across America.
1.1 — The Return of Utility Politics
The empirical record is unambiguous. Before 2019, average U.S. electricity prices had been essentially flat for more than a decade, hovering at roughly 13 cents per kilowatt-hour—a plateau produced by the combination of energy efficiency improvements, cheap natural gas, and stable regulatory environments.3 By the end of 2025, that average had risen to approximately 19 cents per kilowatt-hour, an increase of roughly 27 percent in just six years. This is not merely an inflation story; the consumer price index for the same period reflected a general increase of under 20 percent, meaning electricity has been inflating faster than the broader economy.
The aggregate picture is damaging enough. The disaggregated picture is alarming. Utilities in 34 states requested more than $29 billion in rate increases in the first half of 2025 alone—double the amount requested in the equivalent period of 2024.4 States with high concentrations of data centers have fared considerably worse than the national average. Virginia’s electricity prices surged 13 percent in the relevant period; Illinois, 16 percent; Ohio, 12 percent—all substantially above the national average of 6 percent.5
The human dimension of these increases is not abstract. Total outstanding utility bill debt for U.S. households reached $25 billion in mid-2025, up from approximately $15 billion in early 2022.6 Utility shut-offs climbed to 3.5 million in 2024 and may have reached 4 million in 2025. A 2026 study found that approximately 1 in 6 American households was behind on utility payments.7 And in a November 2025 nationally representative survey of 2,146 adults by Consumer Reports, 78 percent of Americans reported being somewhat or very concerned that new data center construction would raise their energy bills.8
These numbers represent the raw material of voter revolt. Electricity affordability has become, according to recent polling, a source of financial stress for roughly 40 percent of Americans. That is the kind of number that moves elections.
“We just can’t build enough gas-fired power plants to make up the gap. Coal plants once expected to retire are still online, electricity prices are rising, and emissions are beginning to rise.”
— Professor Joseph Aldy, Heinz Professor of the Practice of Environmental Policy, Harvard Kennedy School, February 2026⁶
1.2 — Campaign Messaging for November 2026
The translation of utility bill inflation into electoral messaging is already underway, and its contours reflect the peculiar political moment in which America finds itself: a country simultaneously proud of its AI leadership and increasingly anxious about who is paying for it. The political optics of trillion-dollar corporations receiving subsidized electricity while working-class families cannot afford their monthly bills is precisely the kind of structural grievance that transcends party affiliation.
Democratic candidates in data-center-heavy districts are gravitating toward messaging that combines consumer protection with anti-corporate subsidy arguments. Proposed slogans—“Families Before Server Farms,” “Stop Corporate Power Grabs,” “Electricity for People, Not Just Platforms,” “No More AI Utility Bailouts”—reflect an effort to frame the AI electricity debate as a class conflict rather than a technical infrastructure dispute.
Republican candidates are navigating a more complex terrain. The traditional Republican affinity for corporate economic development and opposition to regulatory expansion puts them in an awkward position when the corporations in question are extracting subsidies from residential ratepayers. The more promising Republican frame centers on energy reliability, domestic generation supremacy, and nuclear expansion—arguments that position the party as pro-energy rather than pro-Big Tech, while avoiding direct confrontation with the hyperscalers whose political donations and economic development promises carry significant weight.
Virginia provides the most instructive recent data point. In the November 2025 gubernatorial election, then-candidate Abigail Spanberger won by 14 percentage points, running in significant part on a platform that promised data centers would “pay their fair share” of electricity costs.9 Nearly three-quarters of Virginia voters, according to a January 2026 survey, blamed data centers for rising electricity costs—a remarkable level of attribution for a technically complex causal chain.10
The 2026 midterm dynamics are therefore not merely ideological. They are arithmetical. PJM Interconnection, which serves more than 65 million people across 13 states and the District of Columbia, is the largest grid in the United States and the one most acutely affected by data center load growth. Every congressional district within PJM’s footprint contains voters who are paying, through their utility bills, for a portion of the infrastructure expansion that hyperscalers are receiving. That is a large constituency with a concrete grievance and a simple narrative.
1.3 — Virginia: Noise, Land Use, and Political Blowback
Nowhere in America has the collision between hyperscale ambition and residential reality been more visceral than in Loudoun County, Virginia—“Data Center Alley,” the region around Ashburn that hosts the densest concentration of data center infrastructure on earth. Loudoun County currently contains 199 operational data centers, with another 117 in various stages of development.11 The physical and acoustic presence of this infrastructure—diesel backup generators, cooling towers, electrical substations—has generated years of noise complaints, land use disputes, and community opposition that have gradually coalesced into organized political resistance.
Dominion Energy, which serves most of Virginia, has proposed a 14 percent rate increase for residential customers in 2026, explicitly citing data center demand growth as a primary driver.12 This represents a profound break from historical pattern: Dominion’s February 2025 proposal was its first base-rate increase since 1992—a 33-year gap—yet it arrived in the same period when data center construction contracts were being signed at a pace Dominion CEO Bob Blue characterized as “unprecedented,” with the company in contract talks for an additional 47 gigawatts of new load as of late 2025.13
Mark Christie of William & Mary’s Center for Energy Law and Policy articulated the core infrastructure dilemma with stark clarity: if all proposed data centers in Virginia are built, they would require nearly tripling Dominion’s all-time peak electricity load, demanding “enormous additions to the energy infrastructure, at costs that will be in the billions of dollars.”14 The entity responsible for absorbing those billions, Christie noted, is not the data center operators. It is the 2.6 million residential customers on Dominion’s system.
1.4 — Memphis: Environmental Justice and the xAI Controversy
If Northern Virginia represents the affluent face of data center political conflict—homeowners in Loudoun County upset about noise and view-sheds—South Memphis represents its environmental justice face. The installation of Elon Musk’s xAI supercomputer complex, “Colossus,” in a predominantly Black, low-income neighborhood that has already been identified as having a cancer risk four times the national average, has become the most galvanizing environmental justice case study in the AI infrastructure debate.15
According to the Southern Environmental Law Center, xAI has been operating dozens of unpermitted methane gas turbines at its Memphis facility without public notice, permits, or air pollution controls, emitting nitrogen oxides and formaldehyde around the clock in quantities that may make xAI the largest industrial source of smog-forming pollutant in Memphis.16 Memphis had already received an “F” grade from the American Lung Association for ozone pollution in 2025, was failing to meet federal standards for smog, and had been named an “asthma capital.” Into this environment, xAI introduced what environmental attorneys described as an “illegal power plant.”17
“By looking to evade clean air laws to operate dirty turbines that emit pollution and known carcinogens, these companies are following a shameful, familiar pattern: asking Black and frontline communities to bear the toxic brunt of ‘innovation.’”
— Abré Conner, NAACP Director of Environmental and Climate Justice, April 2026¹⁸
In February 2026, the Southern Environmental Law Center and Earthjustice filed a lawsuit representing the national NAACP and its Mississippi State Conference, seeking emergency injunctive relief to stop xAI’s operations. A subsequent study found that xAI’s plans to operate more than 40 permanent methane gas turbines at a second facility, Colossus 2, would significantly increase air pollution across North Mississippi and West Tennessee and impose tens of millions of dollars in annual health damages on already-overburdened communities.19
1.5 — Arizona: Water Stress and AI Infrastructure
The Arizona dimension of the data center controversy introduces a resource dimension that transcends electricity: water. The evaporative cooling systems that hyperscale data centers rely upon to prevent their servers from overheating are extraordinarily water-intensive, creating a direct conflict between AI infrastructure and the already precarious hydrology of the American Southwest.
Municipal governments in the Phoenix metropolitan area, which hosts a substantial and growing concentration of data centers, have faced increasingly difficult negotiations with operators over water allocation, with some municipalities beginning to condition approvals on the adoption of air-cooled or closed-loop water systems. The intersection of drought politics, data center water demand, and electricity costs creates a particularly volatile political environment in a state where water is already a generational conflict.
1.6 — The Political Optics Problem
The unifying political dynamic across all of these geographies—Virginia, Tennessee, Arizona, Georgia, Texas, New Jersey—is what political scientists call the asymmetry of concentrated benefits and diffuse costs. Hyperscalers receive focused, negotiated benefits: long-term power purchase agreements at industrial rates, transmission infrastructure built to serve their load, and regulatory environments shaped by their lobbying. The costs, by contrast, are diffused across millions of residential ratepayers who each pay incrementally more and find it difficult to identify a single responsible party.
Voters tolerate infrastructure investment when they perceive shared benefit. They revolt when benefits appear concentrated among wealthy corporations while costs are spread across households that are already struggling with inflation. The political history of American utility regulation is replete with examples of voter uprisings against perceived corporate extraction—from the Progressive Era battles over railroad rates to the 1970s conflicts over nuclear power cost overruns. The AI electricity controversy is beginning to exhibit the structural characteristics of precisely such a political moment.
Section 2: National Grid Stress: NERC’s Warning and the Economics of Load Shock
Understanding why the AI data center buildout represents a fundamentally different class of infrastructure challenge from previous waves of industrial load growth requires engaging seriously with the technical realities of the American grid. This is not merely an economic or political story; the physical architecture of the bulk power system creates constraints and vulnerabilities that no amount of capital spending can instantly resolve, and the pace of hyperscale demand growth is now measurably outrunning the grid’s capacity to adapt.
2.1 — NERC’s May 2026 Level 3 Essential Action Alert
On May 4, 2026, the North American Electric Reliability Corporation issued a Level 3 Essential Actions Alert—the highest-urgency designation in NERC’s three-tier alert system, and one that is rarely deployed.1 The alert was triggered by a documented pattern of events in which customer-initiated load reductions caused 1,000 megawatts or more of computational load to drop off the bulk power system in seconds, generating major grid stability concerns in both the Eastern Interconnection and the Texas (ERCOT) Interconnection.
The mechanism behind these events is both technically significant and politically illuminating. AI data centers are equipped with protection circuits that automatically disconnect from the grid when they detect voltage disturbances, frequency deviations, or other anomalies that could damage their sensitive computing equipment. From the data center’s perspective, this is rational self-protection—losing a GPU cluster to a power quality event is an expensive catastrophe. From the grid’s perspective, the simultaneous disconnection of a gigawatt or more of load in seconds is itself a reliability event, one that can trigger oscillations, voltage instability, and in extreme cases, cascading failures across the bulk power system.
NERC’s alert identified seven Essential Actions that transmission planners, operators, balancing authorities, and reliability coordinators must now implement, covering modeling data requirements, commissioning practices, protection settings, monitoring protocols, and operating procedures.2 All registered entities were required to acknowledge receipt by May 11, 2026, and must submit detailed responses to 33 specific questions by August 3, 2026. NERC also simultaneously proposed a new functional entity type—the “Computational Load Entity”—to formally register hyperscale data centers within the reliability framework, with initial standards expected by end of 2026.20
Critically, NERC’s own prior Level 2 Alert, issued in September 2025, had revealed that “entities generally did not have sufficient processes, procedures, or methods to address emerging computational loads.”21 The escalation to Level 3 reflects NERC’s determination that the industry’s response to the previous warning was inadequate and that the reliability risks posed by computational load growth now constitute an immediate threat to the bulk power system rather than a future planning concern.
2.2 — America’s Load Growth Shock
The IEA’s April 2026 analysis of data center electricity demand confirms that global data center electricity consumption grew by 17 percent in 2025—in line with projections—while electricity consumption specifically from AI-focused data centers surged 50 percent in the same year.22 The IEA’s central projection sees electricity consumption from data centers roughly doubling from 485 terawatt-hours in 2025 to approximately 950 terawatt-hours in 2030, accounting for around 3 percent of total global electricity demand by that date.23
Within the United States, the concentration of this growth is particularly dramatic. According to the IEA’s Global Energy Review 2026, data centers accounted for approximately 50 percent of total electricity demand growth in the United States in 2025—a figure that crystallizes the lopsided relationship between AI infrastructure investment and the national grid.24 U.S. electricity consumption is projected to rise by close to 2 percent per year on average from 2026 to 2030—more than twice the rate of the previous decade—with data center expansion as the primary driver.25
NERC’s own long-term demand forecasts, issued in early 2026, project peak demand to rise 24 percent nationally on the basis of new data center load alone.26 Goldman Sachs has projected a 160 to 165 percent increase in data center power demand by 2030 compared with 2022 baseline levels. By 2028, U.S. data center electricity consumption is projected to reach between 325 and 580 terawatt-hours, with AI workloads driving the bulk of that growth and accounting for more than half of all data center electricity consumption.27
The pace of these projections is without precedent in the modern history of the American grid. Electric utilities are infrastructure businesses designed to plan in decade-long cycles; the AI data center buildout is demanding capacity decisions in years, sometimes months. The result is a structural mismatch between the speed of private AI investment and the speed of public grid adaptation—a mismatch whose costs are not borne by those who created it.
2.3 — Utility Rate Increases Across the Country
The rate increase picture varies significantly by state and utility, but the trend line is consistent and upward. In Virginia, Dominion has proposed a 14 percent residential rate increase for 2026, its first base-rate increase since 1992.12 In Pennsylvania, PPL Electric Utilities reached a March 2026 settlement agreement that, if approved, would raise average residential bills by 4.9 percent to approximately $184 per month, while simultaneously instituting a new tariff specifically designed for data centers and other large load customers.28
The most alarming long-term projection comes from PJM territory. An analysis projects that unless fundamental changes are made to cost allocation mechanisms, PJM consumers will pay an extra $100 billion through 2033 as new data center load exceeds available power supply. By 2035, some projections suggest that Dominion residential customers could face bill increases of up to $255 per month—a figure that would represent an existential affordability crisis for the 2.6 million households in Dominion’s service territory.9
2.4 — The Cost Transfer Mechanism
The mechanics by which data center electricity costs transfer to residential ratepayers are complex but consequential. Understanding this mechanism is essential to understanding both the scale of the problem and the political vulnerability it creates.
When a hyperscaler constructs a new data center and requests grid connection, the utility must build or upgrade the transmission and distribution infrastructure necessary to serve that load. Under most state regulatory frameworks, these capital investments are added to the utility’s “rate base”—the capital assets on which utilities are permitted to earn a regulated return. Those costs are then recovered from all customers, residential and commercial alike, through general rate increases.3
In PJM, the capacity market provides another channel for cost transfer. The capacity auction for the 2025-2026 delivery year resulted in a total bill of $14.7 billion—a 500 percent increase from $2.2 billion in the 2024-2025 auction. An independent watchdog analysis found that data center demand, actual and forecast, accounted for $9.3 billion—or 63 percent—of that total capacity bill. That cost flows directly to residential ratepayers through a mechanism called capacity cost allocation.5
More recently, for the 2026-2027 delivery year, PJM capacity market clearing prices reached $329.17 per megawatt—over ten times higher than the $28.92 per megawatt price in the 2024-2025 delivery year, with rapid data center growth identified as a major contributing factor.29
The crucial asymmetry in this system is temporal and contractual. Hyperscalers negotiate long-term power purchase agreements that provide them with price certainty over decades. Residential customers are exposed to whatever rate adjustments regulators approve in each rate case cycle, with no ability to negotiate, no ability to hedge, and no ability to exit the market. The hyperscaler locks in the economics; the household absorbs the variance.
By end of 2025, utility companies in 34 states had created 65 special electric rates—known as tariffs—for specific large power users, with 30 of them established in 2025 and 2026 alone.30 While these tariffs represent a step toward more equitable cost allocation, critics argue that they do not fully capture the infrastructure investment costs triggered by hyperscale demand, and that the political barriers to aggressive cost assignment remain significant given the economic development arguments that data center operators deploy in state capitals and county commission chambers.
2.5 — When AI Demand Becomes Public Utility Politics
The core political economy question that the American electricity system is now confronting can be stated simply: who pays? Hyperscalers argue, with some statistical support, that their presence generates tax revenue, economic activity, and jobs that benefit communities. They argue, with less persuasive evidence, that their electricity demand helps utilities achieve the scale economies necessary to spread infrastructure costs more broadly. Some analysts at the Institute for Energy Research have suggested that, across 50 states, there is no statistically significant relationship between data center concentration and electricity prices.
The E3 consulting group, in a May 2026 white paper, offered a more nuanced assessment, finding that while individual Amazon data center facilities in some utility territories generated net surplus revenue—approximately $3.4 million per site on average—the aggregate regulatory and capacity market impacts of concentrated load growth in regions like PJM have been clearly negative for residential customers, with data centers attributable to approximately 50 percent of the capacity price increase in the 2024-2025 and 2025-2026 auctions.31
The implication is not that data centers are uniformly harmful to ratepayers, but that the existing regulatory framework is not well-designed to ensure that the costs of accommodating hyperscale demand are borne by those who create them rather than distributed across the entire ratepaying population. That framework failure is now producing political consequences.
Section 3: State-by-State Grid Constraints: America’s Emerging Electricity Fragmentation
The American electricity system is not a single national grid but a patchwork of regional transmission organizations, state-regulated utilities, independent power markets, and vertically integrated monopolies, each operating under its own regulatory philosophy and facing its own version of the hyperscale load challenge. The geography of AI data center construction—concentrated initially in Northern Virginia, spreading to Texas, Georgia, Ohio, Indiana, Arizona, and Nevada—has produced state-by-state variations in rate impacts, regulatory responses, and political reactions that collectively paint a picture of accelerating electricity fragmentation.
3.1 — Virginia (PJM): America’s Data Center Epicenter
Virginia’s position as the global epicenter of data center infrastructure is a function of historical accident compounded by deliberate policy. The concentration of federal government and defense-related data needs in Northern Virginia attracted early cloud infrastructure from Amazon, Microsoft, and Google; the resulting agglomeration of fiber, power, and technical talent created self-reinforcing advantages that drew still more investment. Today, Loudoun County alone hosts 199 operational data centers with 117 more in development.11
Dominion Energy has filed applications for 70 gigawatts of new electricity generation capacity to serve this demand, describing the growth as “unprecedented” to the Virginia State Corporation Commission.14 The transmission infrastructure required to support this buildout is generating PJM congestion charges and transmission expansion costs that flow through to all ratepayers in the PJM footprint, not merely Virginia residents.
The political response in Virginia has been swift and bipartisan in its anxiety, if not always in its proposed solutions. The 2026 General Assembly session produced a flurry of data center reform bills addressing siting requirements, rate classification, tax incentives, and environmental review. Governor-elect Spanberger’s landslide victory on a “pay their fair share” platform has shifted the political environment substantially, and the State Corporation Commission’s rate case proceedings have become forums for organized ratepayer advocacy of a kind rarely seen in utility regulation.
3.2 — New Jersey (PJM): Grid Density Without Grid Slack
New Jersey presents the paradox of a highly developed transmission network without the reserve capacity to easily absorb large new loads. The state’s grid, operating within PJM’s footprint, faces both transmission congestion from its own density and the capacity market price impacts generated by data center load growth throughout the PJM region. New Jersey’s November 2025 elections, like Virginia’s, reflected significant voter anxiety about electricity affordability, with utility cost increases becoming a campaign issue in multiple legislative and congressional races.
3.3 — Texas (ERCOT): The Flexible Grid Experiment
Texas presents the most structurally innovative response to hyperscale data center load growth, reflecting the ERCOT system’s unique design as the only major U.S. grid operating as a largely isolated energy-only market. In 2025, Texas passed Senate Bill 6, which fundamentally restructured the interconnection process for large electrical loads—specifically data centers exceeding 75 megawatts—within the ERCOT grid, requiring demonstrated ability to participate in demand response and curtailment programs as a condition of interconnection.29
ERCOT’s “controllable load resource” framework represents a policy experiment of national significance. By treating hyperscale data centers as assets that can provide grid services through curtailment and demand response, rather than simply passive loads that draw power, ERCOT is attempting to convert a reliability liability into a reliability asset. The framework requires large computational loads to register, maintain curtailment capability, and participate in grid stability programs—conditions that hyperscalers operating in Texas have generally accepted in exchange for access to Texas’s abundant renewable energy and relatively streamlined permitting environment.
3.4 — California (CAISO): AI Demand Meets Electrification
California’s electricity challenge is uniquely complex because it sits at the intersection of three concurrent demand pressures: hyperscale AI data center growth, aggressive vehicle electrification mandates, and building electrification policies that are removing gas appliances from homes and businesses. The California Independent System Operator faces a planning environment in which each of these three load categories is growing simultaneously, in a state where renewable energy’s intermittency creates already-challenging dispatch dynamics.
The extension of Diablo Canyon’s operational license, originally scheduled for retirement in 2025, reflects California’s belated recognition that baseload power is not an optional luxury in a grid absorbing this combination of load growth. The Western market integration discussions, which would tie California’s CAISO more tightly to neighboring state grids, represent an attempt to access the geographic diversity of renewable generation to manage what is becoming an extraordinarily tight supply-demand balance.
3.5 — New York: Moratorium Politics and Grid Bottlenecks
New York’s approach to data center growth has been characterized by tension between the economic development ambitions of state government and the grid reliability concerns of the New York Independent System Operator. Several municipalities have enacted or considered moratoria on new data center construction pending resolution of grid capacity questions. The state’s aggressive renewable energy targets create additional complexity: large new computational loads that require consistent 24/7 power are difficult to serve from an increasingly wind-and-solar-dependent grid without either dedicated backup generation or substantial additional transmission infrastructure.
3.6 — Georgia: Tax Incentives, Water Pressure, and Grid Burden
Georgia has aggressively competed for data center investment through a combination of tax incentives, favorable land costs, and the reliability of Georgia Power’s vertically integrated grid. The state’s water resources, however, are under increasing pressure from data center cooling demands, and the infrastructure costs associated with serving hyperscale loads are becoming visible in Georgia Power’s rate case proceedings. The Southern Company subsidiary has signaled significant capital investment requirements to serve its growing data center customer base, raising questions about cost allocation that Georgia’s regulators are only beginning to address systematically.
3.7 — Tennessee / TVA Region: Load Growth Without Fast Enough Supply
The Tennessee Valley Authority’s federally owned and operated grid faces the xAI Memphis situation as a microcosm of the broader challenge: a massive, rapidly deployed data center with power demands that strain the existing system and raise profound questions about the relationship between AI infrastructure and the communities that host it. The TVA region’s industrial heritage makes it politically receptive to large-load economic development, but the environmental justice dimensions of xAI’s South Memphis operations are now generating national attention and legal proceedings that complicate the simple economic development narrative.
3.8 — MISO States: Industrial Load Growth and Reliability Risk
The Midcontinent Independent System Operator footprint—covering Louisiana, Alabama, Mississippi, Oklahoma, and portions of a dozen other states—is experiencing a wave of large-load interconnection requests driven by the combination of AI data centers, electric vehicle battery manufacturing, and onshored semiconductor production. The MISO region’s transmission infrastructure is older on average than PJM’s and faces particular reliability challenges in accommodating the rapid pace of load growth, especially in the southern portion of the footprint where data center development is accelerating on the basis of land availability, lower labor costs, and favorable climate for certain cooling approaches.
3.9 — Southwest States: Water, Cooling, and the Desert Economics Problem
Arizona, Nevada, and New Mexico share a fundamental constraint that sits above and beyond electricity grid capacity: water scarcity in a warming climate. Evaporative cooling systems, which remain the most cost-effective approach for dissipating the heat generated by dense AI compute clusters, consume millions of gallons of water daily at large facilities. Phoenix-area data centers are facing increasing regulatory and political resistance rooted not in electricity affordability alone but in the intersection of electricity demand, water demand, and drought vulnerability in a region that the Colorado River Compact’s crisis has already sensitized to long-term water allocation politics.
3.10 — Emerging States: The New Hyperscaler Geography
Indiana, Ohio, Pennsylvania, and Utah are emerging as second-wave data center destinations as the traditional hubs—Northern Virginia, the San Francisco Bay Area, Chicago—face interconnection queues measured in years and land scarcity that has made construction timelines impractical for hyperscalers operating on quarterly earnings pressure. NERC’s medium-growth scenario projects that seven additional states—including Nevada, Wyoming, and Indiana—could cross the threshold of 20 percent data-center-driven load growth, even as new capacity builds emerge primarily in Ohio, Pennsylvania, Louisiana, and Mississippi.18
Section 4: The Nuclear Revival: Solving Power Grid Constraints or Creating Private Power Islands?
The nuclear dimension of the AI electricity story is simultaneously the most technologically intriguing and the most politically consequential. The renaissance of nuclear power as a serious commercial proposition—after decades of stagnation following Three Mile Island, Chernobyl, and the economics of cheap natural gas—is being driven not by climate policy, not by government mandate, and not by utility planning but by the electricity procurement decisions of the five largest technology companies in the world. This is a structural novelty in energy history, and its implications extend far beyond the electricity market.
4.1 — Why Gas and Renewables Alone Are Not Enough
The fundamental problem facing hyperscalers’ energy procurement strategies can be stated in three words: intermittency, dispatchability, and scale. Renewable energy—solar and wind—produces electricity only when the sun shines and the wind blows, creating a temporal mismatch with AI compute workloads that run 24 hours a day, 365 days a year. While battery storage can address short-duration gaps, it cannot economically substitute for the week-long periods of low wind and cloud cover that characterize real-world weather patterns.
Natural gas is dispatchable—it can generate electricity on demand—but it carries carbon emissions that hyperscalers have committed to eliminate, exposes corporate electricity costs to gas price volatility, and in the post-2022 environment of geopolitical energy uncertainty carries strategic supply risk. Nuclear power uniquely combines the two characteristics that AI infrastructure demands: reliable 24/7 baseload generation without intermittency, and carbon-free output compatible with corporate net-zero commitments. For hyperscalers, nuclear is not a preference; it is an engineering necessity at scale.
4.2 — The Return of Nuclear Economics
The most dramatic symbol of nuclear’s return was Microsoft’s 20-year, $16 billion power purchase agreement with Constellation Energy to restart Unit 1 of the Three Mile Island nuclear generating station in Pennsylvania—the same plant whose 1979 accident gave American nuclear power its defining public relations catastrophe. Now officially renamed the Christopher M. Crane Clean Energy Center, the 835-megawatt facility is expected to begin delivering power to Microsoft’s operations by 2027, representing the first restart of a retired nuclear reactor in U.S. history.32
The Three Mile Island deal catalyzed a wave of nuclear procurement activity across the hyperscaler peer group. Amazon secured a 17-year, 1.92-gigawatt power purchase agreement with Talen Energy for the Susquehanna nuclear plant, later committing more than $20 billion to transform the surrounding area into an AI campus. Google committed to a 500-megawatt agreement with Kairos Power for advanced small modular reactors. Meta announced a 6.6-gigawatt nuclear procurement strategy, including a January 2026 agreement with Oklo for a 1.2-gigawatt power campus using 16 Aurora Powerhouse reactors in Pike County, Ohio, with first power expected by 2030.33
As of May 2026, all four major hyperscalers have signed at least one nuclear power deal, with 13 announced projects collectively committing more than 9.8 gigawatts of nuclear capacity to AI data center power. The market for nuclear-powered data center sites has developed rapidly, with developers of nuclear-adjacent sites now commanding lease rate premiums of 15 to 25 percent above comparable grid-connected alternatives.34
4.3 — Small Modular Reactors as AI Infrastructure Power
The SMR dimension of the nuclear revival raises questions that extend well beyond electricity economics. Small modular reactors—modular, factory-manufactured nuclear plants designed to generate between 50 and 500 megawatts—are not yet commercially deployed at scale, but hyperscalers are committing capital to them years before they will generate their first electrons. This capital commitment is itself a form of market intervention: by providing anchor tenant offtake agreements, hyperscalers are creating the financing conditions necessary for SMR developers to proceed with projects that would otherwise be unfinanceable.
The critical policy question is whether these SMR investments constitute public infrastructure or captive private compute infrastructure. When a hyperscaler finances a nuclear reactor through a 20-year dedicated power purchase agreement, that reactor is effectively removed from the public grid’s planning horizon. Its power is pre-sold to a single corporate customer; it is not available to reduce capacity prices in the regional market; it does not help the residential ratepayer whose bills are rising because existing generation is insufficient to meet aggregate demand. The reactor exists, economically, as a component of private corporate infrastructure that happens to be licensed by the federal government as a public utility.
The Federal Energy Regulatory Commission’s November 2024 rejection of the proposed behind-the-meter arrangement between AWS and Talen Energy for the Susquehanna nuclear plant is instructive. FERC’s concern was precisely the concern articulated above: allowing a hyperscaler to consume the output of a nuclear plant behind the meter—outside the wholesale market—would raise public power bills by removing a source of capacity from the market while failing to provide any reliability benefit to other grid participants.45
4.4 — Behind-the-Meter Power for Hyperscalers
The Q1 2026 earnings calls of the major hyperscalers revealed the full ambition of the energy sovereignty project that is underway. Meta now expects to spend as much as $145 billion in capital expenditures in 2026—more than double its 2025 capex of $72 billion. Alphabet’s capital expenditures totaled $35.67 billion in Q1 2026 alone, up 106 percent year over year, as part of a full-year plan to spend between $175 and $185 billion. Amazon spent $43.2 billion on capital expenditures in Q1 2026, the overwhelming majority tied to AWS and AI infrastructure.35
Collectively, the four major hyperscalers—Alphabet, Amazon, Meta, and Microsoft—are on track to spend more than $700 billion on capital expenditures in 2026, representing a 77 percent increase from 2025.36 A structural analysis of this spending reveals that the binding constraint is no longer computing hardware—chips are available with sufficient advance commitment—but electricity. As one infrastructure analyst noted, the overwhelming majority of this capital is allocated not to GPUs but to power procurement, generation, and associated grid interconnection.37
Microsoft CEO Satya Nadella acknowledged in early 2026 that Microsoft has GPU clusters sitting idle—depreciating assets waiting for power that may not arrive for years.41 This admission crystallizes the strategic reality: for the world’s most valuable technology companies, the limiting input to AI infrastructure is no longer silicon but electricity.
4.5 — The Risk of Private Energy Sovereignty
The trajectory of hyperscaler energy procurement raises a strategic scenario that this paper identifies as the central long-run risk of the current trajectory: the progressive separation of AI infrastructure from the public grid, creating a two-tier electricity economy in which hyperscalers operate on a portfolio of dedicated nuclear, renewable, and behind-the-meter generation assets while the public grid serves everyone else.
This is not a conspiracy scenario but a structural economic outcome. Hyperscalers have every rational incentive to seek electricity price certainty, carbon-free generation to satisfy corporate sustainability commitments, and reliability assurance that the public grid, under conditions of accelerating load growth and constrained supply, cannot guarantee. The pursuit of these legitimate corporate objectives, through the mechanisms of nuclear PPAs, SMR development, and behind-the-meter generation, produces as a side effect the progressive depletion of the market signals and capacity resources that the public grid depends upon to maintain reliability for all other participants.
“If hyperscalers build dedicated energy ecosystems, public grids become secondary infrastructure—serving the customers that corporations do not find worth serving directly.”
— Power & AI Initiative, Harvard Paulson School of Engineering and Applied Sciences, February 2026³⁵
4.6 — Compute Colonialism in Electricity Markets
The concept of “compute colonialism” in electricity markets captures something important about the emerging hierarchy of access to the electrical grid. At the apex of this hierarchy sit the hyperscalers, whose financial resources allow them to secure long-term power purchase agreements, fund nuclear restarts, and develop behind-the-meter generation capacity that gives them near-total insulation from public grid price volatility. Below them sit large industrial consumers, who have the scale to negotiate special tariffs and demand response contracts but lack the capital to pursue dedicated generation. Below them sit commercial customers, subject to whatever tariff structure the utility proposes and the regulator approves. At the base sit residential households, who have no negotiating power, no alternative, and no hedge against the infrastructure costs generated by the tiers above them.
This hierarchy is not a conspiracy; it is the emergent structure of a deregulated energy economy in which capital resources, contract duration, and technical sophistication translate directly into electricity cost advantage. Its political consequence is the subject of the final section of this paper.
Section 5: Strategic Lessons of Power Grid Constraints
The analysis in the preceding sections converges on five strategic lessons that policymakers, regulators, and citizens must internalize if the American electricity system is to navigate the AI transition without generating the kind of sustained political and economic damage that would ultimately harm the AI industry itself, along with the broader public. These lessons are not partisan positions; they reflect structural realities about infrastructure, political economy, and the social contract that any serious policymaker—Democrat or Republican, progressive or conservative—must engage with.
Pillar I — Electricity Is No Longer Neutral Infrastructure
The first lesson is foundational: electricity has ceased to function as neutral infrastructure and has become strategic allocation infrastructure. For most of the twentieth century, the central regulatory principle of electric utility law was non-discrimination: utilities were required to serve all customers within their service territory at rates that did not unreasonably prefer some customers over others. That principle was never perfectly implemented, but it served as a normative anchor for utility regulation.
The hyperscale AI buildout has rendered this principle increasingly fictional. Long-term power purchase agreements negotiated between trillion-dollar corporations and utilities represent allocations of grid capacity that are effectively unavailable to the residential ratepayers who also depend on that capacity. The regulatory fiction that all customers receive equivalent treatment is being exposed by the structural reality that hyperscalers have converted their financial scale into electricity priority in ways that ordinary households cannot replicate. Regulators and legislators must explicitly acknowledge this shift and design frameworks that honestly address the resulting allocation questions.
Pillar II — AI Converts Compute Demand into Political Demand
The second lesson concerns the political economy of technological infrastructure. When the costs of AI development were primarily borne by venture capitalists and technology company shareholders, the political system had little reason to engage deeply with AI’s resource requirements. The moment those costs began appearing on 130 million household electricity bills, AI infrastructure became a political issue in the most direct and consequential sense.
This is not a temporary aberration that will resolve itself as grid investment catches up with demand. The scale of the projected demand growth—a potential doubling of data center electricity consumption by 2030, concentrated in specific geographies—ensures that the electricity costs of AI will be a persistent feature of the political landscape for at least a decade. The AI industry’s failure to proactively develop regulatory frameworks that address cost equity has left it exposed to exactly the kind of populist political backlash that Virginia’s elections presaged.
Pillar III — Ratepayer Socialism, Shareholder Capitalism
The third lesson is the sharpest conceptual argument in this paper, and it is also the most politically potent: the current structure of AI electricity economics socializes infrastructure risk while privatizing financial returns. Publicly regulated utilities build transmission lines, upgrade substations, and procure generation capacity to serve hyperscale loads. The costs of this investment are distributed across all ratepayers through general rate increases. The financial returns—the cloud computing revenues, the AI service subscription fees, the data center REIT distributions—flow entirely to hyperscaler shareholders, their investors, and the technology ecosystem they support.
This asymmetry is the defining political economy tension of the AI electricity debate, and it is the argument that will resonate most powerfully with the working-class and middle-class voters who are experiencing electricity bill increases while reading about trillion-dollar technology company market capitalizations in the same news cycle. The phrase “ratepayer socialism, shareholder capitalism” captures the structure of this grievance with an accuracy that political operatives will not ignore.
“In regional electricity markets such as PJM, which serves much of the Mid-Atlantic, rapid data-center growth has contributed to higher prices and renewed debates over whether consumers or companies should bear the cost of new infrastructure.”
— Jason Furman, Aetna Professor of the Practice of Economic Policy, Harvard Kennedy School; and Joseph Aldy, Heinz Professor of the Practice of Environmental Policy, Harvard Kennedy School, February 2026³⁹
Pillar IV — Grid Flexibility Must Become Mandatory for Large Computational Loads
The fourth lesson is operational and regulatory rather than political. The NERC Level 3 Alert of May 2026 is not merely a warning about past grid events; it is a design requirement for the future. The current paradigm in which AI data centers receive firm, uninterruptible electricity service while contributing to grid instability through sudden mass disconnection events is not sustainable as computational loads scale to tens and eventually hundreds of gigawatts.
The Texas ERCOT model of treating large computational loads as controllable load resources—requiring registration, curtailment capability, and active participation in grid stability programs—should become a national standard rather than a state experiment. Grid flexibility, demand response participation, and interruptibility should be conditions of interconnection for large computational loads, not voluntary options that hyperscalers may adopt when economically convenient. The NERC reliability guidelines and the proposed “Computational Load Entity” registration framework are steps in this direction, but the pace of implementation must match the pace of deployment.
Pillar V — America Needs an AI-Electricity Compact
The fifth and most comprehensive lesson is that the United States needs a deliberately negotiated AI-Electricity Compact: a policy framework that defines the terms under which hyperscale computational infrastructure can access the public grid while providing enforceable protections for residential ratepayers, environmental justice communities, and grid reliability.
Such a compact should include: mandatory large-load cost responsibility provisions that require hyperscalers to directly fund the transmission and generation investments their load growth necessitates, rather than having those costs socialized through general rate cases; transparent long-term contract disclosure requirements that allow regulators and the public to understand the terms of the electricity deals hyperscalers are negotiating; rigorous environmental review for large load interconnection requests, with specific attention to cumulative air quality impacts in communities already identified as environmental justice concerns; robust consumer protection frameworks that prevent residential ratepayers from bearing the infrastructure costs of industrial customers whose revenues and market capitalizations dwarf those of most sovereign nations; and capacity accountability requirements that tie hyperscaler access to grid capacity to meaningful commitments to provide flexible load response in grid emergency conditions.
The hyperscalers themselves, through their March 2026 White House Pledge, have acknowledged the political necessity of contributing more directly to grid infrastructure costs. The Pledge’s call for large-load rate classes, tighter load forecasts, and a reliability backstop auction mechanism represents a recognition that the current arrangement is politically unsustainable.
The Council on State Governments, backed by the governors of all 13 PJM states, joined Trump’s Energy Dominance Council in a non-binding Statement of Principles urging PJM to run a one-off reliability backstop auction offering 15-year capacity contracts, create large-load rate classes so data centers shoulder more costs, and consider “connect-and-manage” rules under which data center load growth that does not bring associated new supply may face curtailment.18
Conclusion:
This paper began with a simple observation: for decades, electricity in America was invisible infrastructure. It concludes with a more complex and urgent one: electricity is fast becoming the most contested resource in the American political economy.
America’s AI race is not merely a chip race. It is an electricity allocation race. The decisive question is no longer whether the nation can build more compute. NVIDIA’s production lines, Amazon’s GPU clusters, and Google’s custom TPU silicon confirm that compute is being built at a pace that strains every supply chain it touches. The decisive question is whether the social contract surrounding electricity—the implicit agreement between utilities, regulators, and ratepayers that the grid will serve everyone fairly at a price reflecting shared infrastructure investment—can survive the transition that AI is imposing upon it.
The evidence assembled in this paper suggests that the contract is under severe stress. The NERC Level 3 Alert confirms that hyperscale AI data centers are introducing grid instability risks that the existing regulatory framework did not anticipate and cannot easily manage. The PJM capacity market data confirms that data center load growth has already transferred billions of dollars in infrastructure costs to residential ratepayers in 13 states. The Virginia, Tennessee, and Arizona case studies confirm that communities are experiencing the environmental, acoustic, and financial externalities of AI infrastructure without receiving proportionate benefit. The hyperscaler Q1 2026 earnings reports confirm that the companies generating these externalities are doing so in the context of record capital spending and expanding profit margins.
The IEA’s updated projections, confirmed in April 2026, see data center electricity consumption doubling from 485 terawatt-hours in 2025 to approximately 950 terawatt-hours by 2030. If the current cost allocation framework remains unchanged, the residential and commercial ratepayers who make up the vast majority of electricity customers will absorb a growing share of the infrastructure investment required to serve a demand surge from which they derive limited direct benefit.23
The political consequences are already visible and accelerating. Virginia’s 2025 election results, New Jersey’s shifting political environment, the litigation cascading across Memphis, the moratorium proposals emerging from New York to Idaho—these are early signals of a political mobilization that will intensify as electricity bills continue to rise and as ordinary Americans connect that rise to the AI infrastructure they see being built around them.
If ordinary households conclude that the AI economy made billion-dollar infrastructure owners richer while making monthly utility bills less affordable, “Power Grid Constraints” will stop being a technical phrase and become a political slogan. The AI industry, the hyperscalers, and the policymakers who have encouraged this buildout without designing equitable cost frameworks have a narrow window to negotiate the AI-Electricity Compact that can prevent that outcome.
The grid’s constraints are real. The political constraints are tightening. The window for equitable resolution is open. But not indefinitely.
Footnotes and Endnotes:
1. North American Electric Reliability Corporation (NERC). Level 3 Essential Action Alert: Computational Load Modeling, Studies, Instrumentation, Commissioning, Operations, Protection, and Control. May 4, 2026. https://www.nerc.com/globalassets/programs/bpsa/alerts/level-3-computational-load-alert.pdf
2. Morgan Lewis (Power & Pipes). NERC Alert Moves Data Centers From Emerging Risks to Planning Obligations. May 2026. https://www.morganlewis.com/blogs/powerandpipes/2026/05/nerc-alert-moves-data-centers-from-emerging-risks-to-planning-obligations
3. Environmental and Energy Study Institute (EESI). Data Center Power Demands Are Contributing to Higher Energy Bills. February 24, 2026. https://www.eesi.org/articles/view/data-center-power-demands-are-contributing-to-higher-energy-bills
4. Environmental and Energy Study Institute (EESI). Data Center Power Demands Are Contributing to Higher Energy Bills. February 24, 2026. https://www.eesi.org/articles/view/data-center-power-demands-are-contributing-to-higher-energy-bills
5. CNBC. Data centers are concentrated in these states. Here’s what’s happening to electricity prices. November 17, 2025. https://www.cnbc.com/2025/11/14/data-centers-are-concentrated-in-these-states-heres-whats-happening-to-electricity-prices-.html
6. Environmental and Energy Study Institute (EESI). Data Center Power Demands Are Contributing to Higher Energy Bills. February 24, 2026. https://www.eesi.org/articles/view/data-center-power-demands-are-contributing-to-higher-energy-bills
7. The Pew Charitable Trusts. Distributed Energy Can Unleash the Resilient, Affordable Grid of the Future. April 2026. https://www.pew.org/en/research-and-analysis/reports/2026/04/distributed-energy-can-unleash-the-resilient-affordable-grid-of-the-future
8. Consumer Reports. AI Data Centers: Big Tech’s Impact on Electric Bills, Water, and More. March 20, 2026. https://www.consumerreports.org/data-centers/ai-data-centers-impact-on-electric-bills-water-and-more-a1040338678/
9. Introl Blog. Data Center Backlash: $255/Month Bills Swing Elections. January 17, 2026. https://introl.com/blog/data-center-political-backlash-electricity-prices-virginia-december-2025
10. Consumer Reports / Global Strategy Group / Chesapeake Climate Action Network. January 2026 Virginia Voter Survey on Data Centers and Electricity Costs. January 2026. https://www.consumerreports.org/data-centers/ai-data-centers-impact-on-electric-bills-water-and-more-a1040338678/
11. Broadband Breakfast. Dateline Ashburn: Data Centers Drive New Energy Disputes in Northern Virginia. September 11, 2025. https://broadbandbreakfast.com/dateline-ashburn-data-centers-drive-new-energy-disputes-in-northern-virginia/
12. Broadband Breakfast / Patch. Dominion Energy Proposed 14% Rate Increase, Citing Data Center Demand. September 3, 2025. https://new.patch.com/virginia/across-va/dominion-proposes-higher-utility-rates-new-rate-class-data-centers
13. VPM / Virginia Public Media. Virginia Lawmakers Propose Data Center Reform Bills. January 16, 2026. https://www.vpm.org/generalassembly/2026-01-16/2026-data-center-bills-thomas-hb155-mcauliff-hb503-pjm-dominion-energy
14. PCR Homes / William & Mary Center for Energy Law and Policy (Mark Christie). Virginia’s Data Center Boom Faces Growing Community Backlash. May 11, 2026. https://www.pcrehomes.com/blog/virginia-data-center-boom-community-opposition-real-estate-2026/
15. Southern Environmental Law Center (SELC). xAI Built an Illegal Power Plant to Power its Data Center. April 2026. https://www.selc.org/news/xai-built-an-illegal-power-plant-to-power-its-data-center/
16. Tennessee Lookout. A Billionaire, an AI Supercomputer, Toxic Emissions and a Memphis Community That Did Nothing Wrong. July 7, 2025. https://tennesseelookout.com/2025/07/07/a-billionaire-an-ai-supercomputer-toxic-emissions-and-a-memphis-community-that-did-nothing-wrong/
17. Southern Environmental Law Center. Resistance Against xAI Facility in South Memphis Gets Stronger. February 16, 2026. https://www.selc.org/news/resistance-against-elon-musks-xai-facility-in-south-memphis-gets-stronger/
18. Power Magazine. Hyperscalers Sign White House Pledge to Fund Data Center Power, Grid Upgrades. March 6, 2026. https://www.powermag.com/hyperscalers-sign-white-house-pledge-to-fund-data-center-power-grid-upgrades/
19. Southern Environmental Law Center (SELC). New Study Finds Proposed xAI Gas Plant Could Worsen Regional Air Pollution, Cause Millions in Annual Health Damages. February 16, 2026. https://www.selc.org/press-release/new-study-finds-proposed-xai-gas-plant-could-worsen-regional-air-pollution-cause-millions-of-dollars-in-annual-health-damages/
20. Davis Wright Tremaine. NERC Signals Rare Level 3 Alert on Large Loads and Data Centers. May 2026. https://www.dwt.com/blogs/energy–environmental-law-blog/2026/05/nerc-level-3-alert-large-loads-data-centers
21. Utility Dive. Sudden Data Center Load Losses Prompt NERC Alert, Recommendations. April 21, 2026. https://www.utilitydive.com/news/data-center-load-disruptions-nerc-alert-recommendations/818036/
22. IEA (International Energy Agency). Key Questions on Energy and AI – Executive Summary. April 2026. https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary
23. IEA (International Energy Agency). Energy and AI – Energy Demand from AI. 2025-2026. https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
24. IEA (International Energy Agency). Global Energy Review 2026 – Global Trends. April 2026. https://www.iea.org/reports/global-energy-review-2026/global-trends
25. IEA (International Energy Agency). Electricity 2026 – Demand. February 2026. https://www.iea.org/reports/electricity-2026/demand
26. Utility Dive. NERC Forecasts Peak Demand to Rise 24% on New Data Center Loads. January 30, 2026. https://www.utilitydive.com/news/nerc-issues-rare-level-3-alert-over-data-center-load-losses/819295/
27. Scientific American. AI’s Power Needs Will Destroy the Renewable Energy Revolution. May 2026. https://www.scientificamerican.com/article/ais-power-needs-will-destroy-the-renewable-energy-revolution/
28. Utility Dive. PPL Electric Reaches $275M Rate Case Settlement, Including Data Center Tariff. March 16, 2026. https://www.utilitydive.com/news/ppl-electric-rate-case-settlement-data-center-tariff/814760/
29. ArXiv / Harvard Paulson School. Electricity Demand and Grid Impacts of AI Data Centers: Challenges and Prospects. 2026. https://arxiv.org/html/2509.07218v3
30. San.com / Latitude Media. States Take Aim at Data Center Electric Rates. February 26, 2026. https://san.com/cc/states-take-aim-at-data-center-electric-rates-heres-why-it-wont-lower-your-bill/
31. E3 Consulting Group. Are Data Centers Driving Up Electricity Rates? A New E3 Whitepaper Examines the Quantitative Evidence. May 2026. https://www.ethree.com/electricity-rate-drivers-data-center-role-2026/
32. SMR Intel. Every Nuclear-Powered Data Center Deal: Google, Amazon, Meta & Microsoft. May 2026. https://smrintel.com/nuclear-data-center-deals/
33. SMR Data Centers (iRecruit). How Small Modular Reactors Power AI Infrastructure. May 2026. https://www.irecruit.co/insights/smr-nuclear-powered-data-center-developments
34. Build Inc.. Nuclear Power for Data Centers: What the Hyperscaler Procurement Rush Means for Developers. May 2026. https://build.inc/insights/nuclear-power-data-center-development-2026
35. RBN Energy. Q1 2026 Earnings Calls: Meta Ramps Up Spending to Fuel Massive Data Center Buildout. April 29, 2026. https://rbnenergy.com/daily-posts/analyst-insight/q1-2026-earnings-calls-meta-ramps-spending-fuel-massive-data-center
36. Fortune / Eye on AI. Big Tech Is About to Spend $700 Billion on AI This Year. April 30, 2026. https://fortune.com/2026/04/30/big-tech-hyperscalers-will-spend-700-billion-on-ai-infrastructure-this-year-with-no-clear-end-in-sight-eye-on-ai/
37. Next Waves Insight. Hyperscaler Capex 2026: Where the $300 Billion in AI Infrastructure Is Actually Going. 2026. https://nextwavesinsight.com/hyperscaler-ai-capex-microsoft-google-amazon-meta-2026/
38. Harvard Magazine. AI Is Risky Business for the Power Grid, Harvard Experts Say. February 19, 2026. https://www.harvardmagazine.com/harvard-kennedy-school-of-government/harvard-policymakers-investment-ai-risk-energy
39. Belfer Center for Science and International Affairs, Harvard University. AI, Data Centers, and the U.S. Electric Grid: A Watershed Moment. February 10, 2026. https://www.belfercenter.org/research-analysis/ai-data-centers-us-electric-grid
40. Carbon Direct. Inside NERC’s Level 3 Alert on Data Center Loads. May 2026. https://www.carbon-direct.com/insights/nerc-level-3-alert-data-center-loads
41. World Economic Forum / Stanford Research. AI Doesn’t Need More Power, It Needs a Smarter Energy Grid. March 31, 2026. https://www.weforum.org/stories/2026/03/ai-needs-a-smarter-energy-grid/
42. NERC Large Loads FAQ. Large Loads Frequently Asked Questions. May 2026. https://www.nerc.com/globalassets/initiatives/large-loads-action-plan/large-loads-faqs.pdf
43. Earthjustice. NAACP Asks Court for Emergency Action to Stop Illegal Air Pollution from xAI’s Data Center Power Plant. May 2026. https://earthjustice.org/press/2026/naacp-asks-court-for-emergency-action-to-stop-illegal-air-pollution-from-xais-data-center-power-plant
44. Data Center Knowledge. 2026 Predictions: AI Sparks Data Center Power Revolution. April 17, 2026. https://www.datacenterknowledge.com/operations-and-management/2026-predictions-ai-sparks-data-center-power-revolution
45. Data Center Dynamics. Three Mile Island Nuclear Plant Restart Ahead of Schedule. June 27, 2025. https://www.datacenterdynamics.com/en/news/three-mile-island-nuclear-plant-restart-ahead-of-schedule-in-boon-to-microsofts-ai-ambitions-report/
